Error while adding custom index using IndexIDMap

[ravitashaw]

Hi, I'm facing difficulty in adding custom index (filename) to the IndexMap. The objective of the task is to add feature vectors for indexing and while searching the output should be filename rather than an ordinal index. The test code is:

def test_faiss():
    d = 20
    nb = 1
    file_name = 'test.png'
    xb = np.random.random((nb, d)).astype('float32')
    ids = np.array(file_name).reshape(nb, 1)
    index = faiss.IndexFlatL2(d)
    index = faiss.IndexIDMap(index)
    index.add_with_ids(xb, ids)
    return

and, the error is:

assert ids.shape == (n, ), 'not same nb of vectors as ids'
AssertionError: not same nb of vectors as ids

I'm using only 1 vector for indexing, so my ids also has shape of (1, 1). Is there something missing or misunderstood?

[ravitashaw]

I figured out from #641

The above code works fine when the ids is replaced as an integer array. The custom index mapping need to be manually managed.

But the error seems misleading, the shape of ids is ok but the datatype.

[SlytherinWyne]

I'm encountering the same issue, the dimension of xb and ids seem can not be 1, quite weird, annoying error...